Minutes is a Python library for speaker diarization: the process of splitting up an audio file into its unique speakers. Its base model is trained on hundreds of hours of spoken English text, and it uses transfer learning to predict on new speakers.
Minutes is built on Keras running on top of TensorFlow.
Figuring out where and what to eat when you're out with your friends or family can be a daunting task. Eat is an application built to help you decide on a restaurant in your area that accommodates everyone's preferences. Our algorithm suggests places to eat based the group's location, dietary restrictions, price range, and intended meal time. After each person in the group as voted on the suggested restaurants the app selects the optimal restaurant for the group.
Eat was built for iOS using Swift and Objective-C.
Inertia is a cross-platform command line tool that simplifies setup and management of automated project deployment on any virtual private server. It aims to provide the ease and flexibility of services like Heroku without the complexity of Kubernetes while still giving users full control over their projects.
Inertia features a daemon, a web front-end, and a command-line interface. The backend and CLI are built in Go and use Docker, while the frontend is built on React.
Etch is a live multi-user collaborative etch-a-sketch built on websockets. It also features a live chat that users can join anonymously or with a custom username. Each user is assigned their own color and can draw whatever their imagination allows while watching others draw at the same time.
Etch was built using React and WebSockets on the frontend and Node.js on the server side.
Cumulus is a new cryptocurrency with its own blockchain and token. A command line utility allows users to send and receive funds securely from their account. Users can also use the Cumulus app to participate in securing the network by mining new tokens for themselves.
Cumulus uses proof-of-work based on SHA256 hashes to secure the network and ECDSA for wallet keys. The client is an open-source Go project.
Driver is a ride-sharing app for Android that ensures everyone gets home safely. Users provide their destination address and Driver intelligently matches passengers to drivers to minimize driving time. When passengers are matched to drivers, Driver generates and displays an efficient route to each passenger's destination.
The Driver app is built with Java and leverages a Firebase backend. Because of this design, the algorithm matching passengers to drivers is distributed across devices without compromising battery life.
Launch Pad's Machine Learning team participated in the annual Data Science Bowl, hosted by Kaggle.com. The task was to predict presence of lung cancer in CT scans in order to create a model for early lung cancer detection. In the United States, lung cancer affects 225,000 people every year, and accounts for $12 billion in health care costs.
The team used several deep learning techniques including convolutional neural networks with several different architectures. Our approach placed in the top 15% of teams.
Sift is a service that gives businesses a birds-eye view into their customers' feedback. Sift connects to feedback forms, support email, and reviews to create a visual summary of a business's customer relationships using natural language processing. This provides a quick way to see everything important that customers are saying.
Sift is built using a React/Redux frontend and a microservice-based backend with services built using Python and Go. Data is processed using a queueing system based on RabbitMQ that delivers tasks to a Celery processing cluster.
Cashout is a fantasy currency trading app. When you open an account, you receive $1000 USD, from there it's up to you! Trade world currencies and cryptocurrencies, and see how your portfolio performs against the USD or world economy indexes.
Cashout was built using Ruby on Rails and Redis. It integrates with popular currency and cryptocurrency price indexes for real-time trading data.
RocketCast is a full-featured podcast player and discovery tool for iOS. Search for and subscribe to podcasts in the world's largest podcast directory, read show notes inline with podcast metadata, and download in-app for offline listening. Skip ads and play at double or triple speed.
RocketCast is built for iOS using Swift and uses the iTunes podcast registry for podcast discovery.
Developed as a replacement for iClickers without the need for specialized hardware, RocketPoll allows users to create and respond to polls in real-time. You can choose custom questions and answers, and limit the time during which questions are valid.
RocketPoll was built using Swift and a FireBase backend.